QuantumATK Forum
QuantumATK => General Questions and Answers => Topic started by: bubble on October 24, 2017, 16:58
-
Hi,
recently, I investigated a 1D nanodevice with spin-polarized calcualation. The attachment is three files. You can see that for the down spin, the average T(E) and the k-resovled transmision coiffiences is of 10^(-6) order of magnitude (the first figure). Therefore, the IV show that Idown is nearly zero (the third figure). However, the calculated the eigenvalue of the spin-down channel is about 0.38 (see the left-down cornor of the first figure) and the corresponding states (the second figure) are delocalized.
I want to know whether it is a bug? OR, how to explain this strange phenominon?
THANK YOU
-
The eigenvalues are related to the electron transmission at the Gamma k-point in the 2D Brillouin zone, whereas what you call the average T(E) is calculated as an integral over the 2D Brillouin zone for a given energy. If the transmission is small at all the k-points except the Gamma-point within the bias window, you may expect that the current will be low. So, I see no contradiction in that regard.
-
It's a 1D system, Petr, not 2D.
The eigenvalues should add up to the transmission, at any given k-point, so there is definitely something odd going on.
-
@bubble Please post the actual script and log file related to this calculation. What version of ATK and operating system are you using?
-
@bubble Please post the actual script and log file related to this calculation. What version of ATK and operating system are you using?
For the py file, please see the attachment. The log file is too large to upload, thus I only paste some useful information (see below)
+------------------------------------------------------------------------------+
| |
| Atomistix ToolKit 2016.3 [Build 7eb51b6] |
| |
+------------------------------------------------------------------------------+
+------------------------------------------------------------------------------+
| |
| Device DFT Calculation [Started Mon Mar 06 18:39:30 2017] |
| |
+------------------------------------------------------------------------------+
+------------------------------------------------------------------------------+
| |
| CPU Information |
| |
+------------------------------------------------------------------------------+
| Process ID 0 at Caculation |
+------------------------------------------------------------------------------+
+------------------------------------------------------------------------------+
| |
| Size of dense matrices = 342 x 342 [2 MB per matrix] |
| |
+------------------------------------------------------------------------------+
|--------------------------------------------------|
Calculating Kinetic Matrix : ==================================================
|--------------------------------------------------|
Calculating Nonlocal Part : ==================================================
+------------------------------------------------------------------------------+
--------------------------------------------------------------------------------
Density Matrix : 316916.52 s 309.49 s 52.07% |=============|
Exchange-Correlation : 237723.35 s 215.92 s 39.06% |==========|
Real Space Integral : 22425.83 s 20.37 s 3.68% ||
Parallel Mixing : 10362.81 s 9.57 s 1.70% |
Transmission Spectrum : 7524.17 s 684.02 s 1.24% |
Valence Density : 4493.48 s 4.08 s 0.74% |
Hartree Potential : 3758.31 s 3.41 s 0.62% |
Diagonalization : 2131.62 s 36.13 s 0.35% |
Setting Density Matrix : 2028.00 s 507.00 s 0.33% |
Real Space Basis : 59.85 s 3.33 s 0.01% |
Constant Terms : 50.46 s 4.21 s 0.01% |
Difference Density : 20.52 s 0.31 s 0.00% |
Core Density : 6.88 s 0.10 s 0.00% |
Neutral Atom Potential : 3.93 s 0.56 s 0.00% |
Loading Modules + MPI : 2.42 s 2.42 s 0.00% |
Basis Set Generation : 2.40 s 0.13 s 0.00% |
File IO, nlsave : 1.94 s 0.08 s 0.00% |
Hubbard Term : 0.02 s 0.00 s 0.00% |
Fixed Spins Term : 0.00 s 0.00 s 0.00% |
--------------------------------------------------------------------------------
Total : 608594.55 s (7d01h03m14.55s)
-
It's a 1D system, Petr, not 2D.
The eigenvalues should add up to the transmission, at any given k-point, so there is definitely something odd going on.
Thank you for your reply, Blom.
Now I think I can confirm that the issue is a bug, please see the figures in the doc file in the attachment.
-
What about operating system? I have noticed that you are using the 2016.3 version. It is always a good idea of using an updated version, in your case this would be 2016.4.
-
What about operating system? I have noticed that you are using the 2016.3 version. It is always a good idea of using an updated version, in your case this would be 2016.4.
It seems that at that time, Mar, 2017, we don't update ATK to 2016.4(or this version is still not available at that time?). But now I have decided to update it to 2017version :D
-
We didn't modify anything in IVCurve from 2016.3 to 2016.4, so let's spare bubble having to rerun the 7-day calculation :)
We have IVCurve objects in our test database, I'm pretty sure we can replicate the error using that.
This is clearly a bug, I have checked the source code. The "matching_configuration" is determined from the IVCurve labfloor item, not the transmission spectrum, which probably means the zero-bias configuration, no matter which transmission spectrum you load. I suspect you could confirm this, bubble, by computing the eigenstates at the same energy for zero bias, they should match using both approaches.
As such I'm sure the problem is isolated to the TransmissionAnalyzer, so bubble I suggest you compute your eigenstates using plain scripts for now.
-
Btw, the bug would be present in 2017 too
-
@bubble Thanks a lot for your report. I will create an internal bug report on this. It would also be helpful to have your output nc-files and full log file. I guess the files are quite bulky. So, could you then put it in Dropbox and send us a link? You may send the links through our support channel, [email protected], if you do not feel like sharing the links publicly.
-
@bubble Thanks a lot for your report. I will create an internal bug report on this. It would also be helpful to have your output nc-files and full log file. I guess the files are quite bulky. So, could you then put it in Dropbox and send us a link? You may send the links through our support channel, [email protected], if you do not feel like sharing the links publicly.
The link to the py and log files has been sent to the 'support' email.
-
The "matching_configuration" is determined from the IVCurve labfloor item, not the transmission spectrum, which probably means the zero-bias configuration, no matter which transmission spectrum you load.
I confirm that. So, this is a bug in Transmission Analyzer.